Digital-first primary care is now firmly embedded across Europe and beyond. Video consultations and patient-uploaded photographs have become routine in triage, diagnosis and follow-up. In many settings, the smartphone has effectively become a front-line diagnostic device. But… smartphones are not clinical instruments.
As reliance on remote visual assessment grows, so too does concern that the images clinicians depend on may not accurately represent clinical reality. Suboptimal lighting, poor framing, low resolution and bandwidth compression can all degrade image quality. More subtly, and potentially more dangerously, software-driven enhancements, including AI-based filters, may alter colour, texture and contrast in ways that obscure clinically relevant signs. The implications for patient safety are significant, according to new research from The Conversation and Bangor University, published in The Lancet Primary Care.
Optimised for beauty, not for medicine
Modern smartphone cameras are engineered to produce visually appealing content. Automatic white balance, dynamic range optimisation and skin-smoothing algorithms are calibrated to create images that look good on social media and messaging platforms. They are not designed to preserve clinical accuracy.
Every step in the capture–transmission–display chain introduces the possibility of distortion. Cameras interpret light through proprietary algorithms. Video platforms compress files to reduce bandwidth. Screens vary widely in colour calibration and brightness. Night modes can introduce yellow tints; compression can blur fine details; automatic enhancement may suppress or exaggerate colour cues.
For conditions in which colour is diagnostic, such as jaundice, cyanosis, pallor or erythema, these shifts matter. A patient with anaemia may appear less pale. A patient with hepatitis may look less jaundiced. Subtle cyanosis or early oedema may be missed altogether. Clinicians may not fully appreciate how extensively consumer devices modify images. The resulting visual data can appear convincing while being technically inaccurate.
Disproportionate impact and subtle bias
These risks are not evenly distributed. Assessing colour-dependent clinical signs is already more complex in individuals with darker skin tones. If imaging algorithms are not optimised across diverse populations, distortions may further reduce diagnostic reliability and widen disparities.
Subtle findings are particularly vulnerable. Mild rashes, faint bruising or early inflammatory changes can be smoothed away by compression or enhancement processes. When neither patient nor clinician recognises that distortion has occurred, visual information may be given unwarranted diagnostic weight. The danger lies less in dramatic error than in misplaced confidence.
The generative AI layer
The integration of generative AI into consumer imaging introduces a new category of risk. Platforms such as Snapchat popularised digital lenses that smooth skin and enhance facial features. Today’s AI systems go further, regenerating facial regions, altering shadows and adjusting skin tone in highly realistic ways.
In everyday life these features are benign or even desirable. In clinical contexts they may unintentionally erase medically relevant cues. A bruise can be softened, asymmetry minimised, discolouration corrected. Increasingly, such enhancements are applied automatically and invisibly.
Patients may not realise filters are active. Clinicians may not suspect they are present. Yet the image on screen may be a computationally reconstructed interpretation rather than a faithful representation.
A gap in standards and awareness
Patient-safety concerns in remote consultations are well recognised. However, comparatively little attention has been paid to the imaging pipeline itself as a source of risk. There are currently no widely adopted minimum standards defining when patient-generated images are of sufficient quality for clinical decision-making. Few healthcare systems provide structured training on recognising technology-induced distortion. Guidance on when to escalate to in-person assessment because of visual uncertainty is limited.
Moreover, practical barriers hinder research. Video consultations are often not routinely stored because of file size constraints, restricting retrospective safety analysis. Interdisciplinary collaboration between clinicians and computer scientists remains the exception rather than the rule. Digital adoption has outpaced digital governance.
Mitigation: pragmatic safeguards now, structural solutions later
Some risk-reduction measures are straightforward. Patients can be encouraged to use natural daylight, avoid mixed lighting, position cameras steadily and ensure that the relevant body area is clearly visible. Explicit prompts to disable beauty modes and filters before capturing images or joining video consultations could become routine. A simple verbal or electronic confirmation that no filters are active may help raise awareness.
Clinicians, for their part, need to approach patient-generated images as potentially distorted representations rather than objective evidence. Embedding prompts into triage templates, asking about lighting conditions or filter use, can normalise this awareness. Confirming impressions directly with patients (“Does this look like your usual skin colour?”) introduces a shared safety check. When image quality is suboptimal or subtle colour-dependent signs are clinically important, the threshold for in-person assessment should remain low.
At system level, deeper changes are required. Research is needed to quantify how consumer imaging pipelines affect the visibility of specific clinical signs. Platforms used in healthcare could provide real-time alerts about poor lighting or excessive compression. Smartphone manufacturers might develop a dedicated “healthcare mode” that minimises enhancement and prioritises colour accuracy. Minimum standards for clinical display screens and transparency in image processing could form part of future regulatory frameworks.
As generative AI becomes more deeply embedded in consumer devices, telehealth systems may also require content-authentication tools to preserve the integrity of clinical observations.
Recalibrating digital care
Smartphones have become ubiquitous tools in remote primary care, but they remain uncalibrated diagnostic instruments. The assumption that visual information transmitted through consumer technology is inherently reliable is no longer defensible.
Visual fidelity must become a recognised patient-safety issue. That means raising awareness among clinicians and patients, embedding pragmatic safeguards into workflows, investing in interdisciplinary research and developing regulatory standards aligned with the realities of AI-enhanced imaging.
Digital care has delivered accessibility and convenience at scale. Ensuring that what clinicians see truly reflects what is clinically present is the next frontier in making remote healthcare not only efficient but safe.
